What is redundant encoding?

Study for the AI, Business Strategy, and Ethics Exam. Prepare with multiple choice questions and comprehensive explanations. Boost your exam confidence with our expertly curated content!

Multiple Choice

What is redundant encoding?

Explanation:
Redundant encoding means the protected attribute isn’t used directly, but signals about it leak into the model through features that are correlated with it. Because those features carry information about the protected attribute, the model can infer and use that information anyway, reintroducing bias even if you didn’t feed the protected attribute itself. That’s why describing a protected attribute being correlated with other features, allowing indirect use, is the correct idea. If there’s no correlation, or if the attribute is simply ignored or included directly, you don’t get this indirect leakage.

Redundant encoding means the protected attribute isn’t used directly, but signals about it leak into the model through features that are correlated with it. Because those features carry information about the protected attribute, the model can infer and use that information anyway, reintroducing bias even if you didn’t feed the protected attribute itself. That’s why describing a protected attribute being correlated with other features, allowing indirect use, is the correct idea. If there’s no correlation, or if the attribute is simply ignored or included directly, you don’t get this indirect leakage.

Subscribe

Get the latest from Examzify

You can unsubscribe at any time. Read our privacy policy